Sincerity detector threatens to expose those little white lies

Students at the Massachusetts Institute of Technology rarely sit still.

Some, like Anmol Madan, a PhD candidate at the MIT Media Lab, occasionally invent something you might expect to find on a late-night TV shopping channel, alongside a do-it-yourself hair transplant kit.

Madan describes his new project, the Jerk-O-Meter (or JerkoMeter), as "a real-time speech feature analysis application that runs on your VOIP phone or cell phone. . . using speech features that measure activity and stress [and soon empathy] from your tone of voice and speaking style, to predict if you are 'being a jerk' on the telephone."

The clunky, hand-held device currently analyses only your conversation, not that of the person you are calling. It transmits messages that range from "Stop being a jerk!" to "Wow, you're a smooth talker", depending on how well you are doing. It can also inform the person on the other end of the line that you are extremely busy. A nice touch.

The mathematical models for the Jerk-O-Meter grew out of several research studies at MIT which evaluated how a speaking style could reflect a person's interest in, say, going on a date or in buying a particular product.

"Our results show that a person's speaking style and tone of voice can predict objective outcomes, for example, interest in a conversation or in going out on a date, with 75 to 80 per cent accuracy," Madan explains. That monitoring technology has far broader applications in areas such as consumer research and marketing.

Madan and his mentor, Dr Alex Pentland, have a broader vision of computers and mobile phones, which "may now understand people better, just as other people do". A depressing thought if ever there was one. - Anna Mundow